/*      $NetBSD: ntp_calendar.c,v 1.12 2024/08/18 20:47:13 christos Exp $       */

/*
* ntp_calendar.c - calendar and helper functions
*
* Written by Juergen Perlinger ([email protected]) for the NTP project.
* The contents of 'html/copyright.html' apply.
*
* --------------------------------------------------------------------
* Some notes on the implementation:
*
* Calendar algorithms thrive on the division operation, which is one of
* the slowest numerical operations in any CPU. What saves us here from
* abysmal performance is the fact that all divisions are divisions by
* constant numbers, and most compilers can do this by a multiplication
* operation.  But this might not work when using the div/ldiv/lldiv
* function family, because many compilers are not able to do inline
* expansion of the code with following optimisation for the
* constant-divider case.
*
* Also div/ldiv/lldiv are defined in terms of int/long/longlong, which
* are inherently target dependent. Nothing that could not be cured with
* autoconf, but still a mess...
*
* Furthermore, we need floor division in many places. C either leaves
* the division behaviour undefined (< C99) or demands truncation to
* zero (>= C99), so additional steps are required to make sure the
* algorithms work. The {l,ll}div function family is requested to
* truncate towards zero, which is also the wrong direction for our
* purpose.
*
* For all this, all divisions by constant are coded manually, even when
* there is a joined div/mod operation: The optimiser should sort that
* out, if possible. Most of the calculations are done with unsigned
* types, explicitely using two's complement arithmetics where
* necessary. This minimises the dependecies to compiler and target,
* while still giving reasonable to good performance.
*
* The implementation uses a few tricks that exploit properties of the
* two's complement: Floor division on negative dividents can be
* executed by using the one's complement of the divident. One's
* complement can be easily created using XOR and a mask.
*
* Finally, check for overflow conditions is minimal. There are only two
* calculation steps in the whole calendar that potentially suffer from
* an internal overflow, and these are coded in a way that avoids
* it. All other functions do not suffer from internal overflow and
* simply return the result truncated to 32 bits.
*/

#include <config.h>
#include <sys/types.h>

#include "ntp_types.h"
#include "ntp_calendar.h"
#include "ntp_stdlib.h"
#include "ntp_fp.h"
#include "ntp_unixtime.h"

#include "ntpd.h"

/* For now, let's take the conservative approach: if the target property
* macros are not defined, check a few well-known compiler/architecture
* settings. Default is to assume that the representation of signed
* integers is unknown and shift-arithmetic-right is not available.
*/
#ifndef TARGET_HAS_2CPL
# if defined(__GNUC__)
#  if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
#   define TARGET_HAS_2CPL 1
#  else
#   define TARGET_HAS_2CPL 0
#  endif
# elif defined(_MSC_VER)
#  if defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM)
#   define TARGET_HAS_2CPL 1
#  else
#   define TARGET_HAS_2CPL 0
#  endif
# else
#  define TARGET_HAS_2CPL 0
# endif
#endif

#ifndef TARGET_HAS_SAR
# define TARGET_HAS_SAR 0
#endif

#if !defined(HAVE_64BITREGS) && defined(UINT64_MAX) && (SIZE_MAX >= UINT64_MAX)
# define HAVE_64BITREGS
#endif

/*
*---------------------------------------------------------------------
* replacing the 'time()' function
*---------------------------------------------------------------------
*/

static systime_func_ptr systime_func = &time;
static inline time_t now(void);


systime_func_ptr
ntpcal_set_timefunc(
       systime_func_ptr nfunc
       )
{
       systime_func_ptr res;

       res = systime_func;
       if (NULL == nfunc)
               nfunc = &time;
       systime_func = nfunc;

       return res;
}


static inline time_t
now(void)
{
       return (*systime_func)(NULL);
}

/*
*---------------------------------------------------------------------
* Get sign extension mask and unsigned 2cpl rep for a signed integer
*---------------------------------------------------------------------
*/

static inline uint32_t
int32_sflag(
       const int32_t v)
{
#   if TARGET_HAS_2CPL && TARGET_HAS_SAR && SIZEOF_INT >= 4

       /* Let's assume that shift is the fastest way to get the sign
        * extension of of a signed integer. This might not always be
        * true, though -- On 8bit CPUs or machines without barrel
        * shifter this will kill the performance. So we make sure
        * we do this only if 'int' has at least 4 bytes.
        */
       return (uint32_t)(v >> 31);

#   else

       /* This should be a rather generic approach for getting a sign
        * extension mask...
        */
       return UINT32_C(0) - (uint32_t)(v < 0);

#   endif
}

static inline int32_t
uint32_2cpl_to_int32(
       const uint32_t vu)
{
       int32_t v;

#   if TARGET_HAS_2CPL

       /* Just copy through the 32 bits from the unsigned value if
        * we're on a two's complement target.
        */
       v = (int32_t)vu;

#   else

       /* Convert to signed integer, making sure signed integer
        * overflow cannot happen. Again, the optimiser might or might
        * not find out that this is just a copy of 32 bits on a target
        * with two's complement representation for signed integers.
        */
       if (vu > INT32_MAX)
               v = -(int32_t)(~vu) - 1;
       else
               v = (int32_t)vu;

#   endif

       return v;
}

/*
*---------------------------------------------------------------------
* Convert between 'time_t' and 'vint64'
*---------------------------------------------------------------------
*/
vint64
time_to_vint64(
       const time_t * ptt
       )
{
       vint64 res;
       time_t tt;

       tt = *ptt;

#   if SIZEOF_TIME_T <= 4

       res.D_s.hi = 0;
       if (tt < 0) {
               res.D_s.lo = (uint32_t)-tt;
               M_NEG(res.D_s.hi, res.D_s.lo);
       } else {
               res.D_s.lo = (uint32_t)tt;
       }

#   elif defined(HAVE_INT64)

       res.q_s = tt;

#   else
       /*
        * shifting negative signed quantities is compiler-dependent, so
        * we better avoid it and do it all manually. And shifting more
        * than the width of a quantity is undefined. Also a don't do!
        */
       if (tt < 0) {
               tt = -tt;
               res.D_s.lo = (uint32_t)tt;
               res.D_s.hi = (uint32_t)(tt >> 32);
               M_NEG(res.D_s.hi, res.D_s.lo);
       } else {
               res.D_s.lo = (uint32_t)tt;
               res.D_s.hi = (uint32_t)(tt >> 32);
       }

#   endif

       return res;
}


time_t
vint64_to_time(
       const vint64 *tv
       )
{
       time_t res;

#   if SIZEOF_TIME_T <= 4

       res = (time_t)tv->D_s.lo;

#   elif defined(HAVE_INT64)

       res = (time_t)tv->q_s;

#   else

       res = ((time_t)tv->d_s.hi << 32) | tv->D_s.lo;

#   endif

       return res;
}

/*
*---------------------------------------------------------------------
* Get the build date & time
*---------------------------------------------------------------------
*/
int
ntpcal_get_build_date(
       struct calendar * jd
       )
{
       /* The C standard tells us the format of '__DATE__':
        *
        * __DATE__ The date of translation of the preprocessing
        * translation unit: a character string literal of the form "Mmm
        * dd yyyy", where the names of the months are the same as those
        * generated by the asctime function, and the first character of
        * dd is a space character if the value is less than 10. If the
        * date of translation is not available, an
        * implementation-defined valid date shall be supplied.
        *
        * __TIME__ The time of translation of the preprocessing
        * translation unit: a character string literal of the form
        * "hh:mm:ss" as in the time generated by the asctime
        * function. If the time of translation is not available, an
        * implementation-defined valid time shall be supplied.
        *
        * Note that MSVC declares DATE and TIME to be in the local time
        * zone, while neither the C standard nor the GCC docs make any
        * statement about this. As a result, we may be +/-12hrs off
        * UTC.  But for practical purposes, this should not be a
        * problem.
        *
        */
#   ifdef MKREPRO_DATE
       static const char build[] = MKREPRO_TIME "/" MKREPRO_DATE;
#   else
       static const char build[] = __TIME__ "/" __DATE__;
#   endif
       static const char mlist[] = "JanFebMarAprMayJunJulAugSepOctNovDec";

       char              monstr[4];
       const char *      cp;
       unsigned short    hour, minute, second, day, year;
       /* Note: The above quantities are used for sscanf 'hu' format,
        * so using 'uint16_t' is contra-indicated!
        */

#   ifdef DEBUG
       static int        ignore  = 0;
#   endif

       ZERO(*jd);
       jd->year     = 1970;
       jd->month    = 1;
       jd->monthday = 1;

#   ifdef DEBUG
       /* check environment if build date should be ignored */
       if (0 == ignore) {
           const char * envstr;
           envstr = getenv("NTPD_IGNORE_BUILD_DATE");
           ignore = 1 + (envstr && (!*envstr || !strcasecmp(envstr, "yes")));
       }
       if (ignore > 1)
           return FALSE;
#   endif

       if (6 == sscanf(build, "%hu:%hu:%hu/%3s %hu %hu",
                       &hour, &minute, &second, monstr, &day, &year)) {
               cp = strstr(mlist, monstr);
               if (NULL != cp) {
                       jd->year     = year;
                       jd->month    = (uint8_t)((cp - mlist) / 3 + 1);
                       jd->monthday = (uint8_t)day;
                       jd->hour     = (uint8_t)hour;
                       jd->minute   = (uint8_t)minute;
                       jd->second   = (uint8_t)second;

                       return TRUE;
               }
       }

       return FALSE;
}


/*
*---------------------------------------------------------------------
* basic calendar stuff
*---------------------------------------------------------------------
*/

/*
* Some notes on the terminology:
*
* We use the proleptic Gregorian calendar, which is the Gregorian
* calendar extended in both directions ad infinitum. This totally
* disregards the fact that this calendar was invented in 1582, and
* was adopted at various dates over the world; sometimes even after
* the start of the NTP epoch.
*
* Normally date parts are given as current cycles, while time parts
* are given as elapsed cycles:
*
* 1970-01-01/03:04:05 means 'IN the 1970st. year, IN the first month,
* ON the first day, with 3hrs, 4minutes and 5 seconds elapsed.
*
* The basic calculations for this calendar implementation deal with
* ELAPSED date units, which is the number of full years, full months
* and full days before a date: 1970-01-01 would be (1969, 0, 0) in
* that notation.
*
* To ease the numeric computations, month and day values outside the
* normal range are acceptable: 2001-03-00 will be treated as the day
* before 2001-03-01, 2000-13-32 will give the same result as
* 2001-02-01 and so on.
*
* 'rd' or 'RD' is used as an abbreviation for the latin 'rata die'
* (day number).  This is the number of days elapsed since 0000-12-31
* in the proleptic Gregorian calendar. The begin of the Christian Era
* (0001-01-01) is RD(1).
*/

/*
* ====================================================================
*
* General algorithmic stuff
*
* ====================================================================
*/

/*
*---------------------------------------------------------------------
* fast modulo 7 operations (floor/mathematical convention)
*---------------------------------------------------------------------
*/
int
u32mod7(
       uint32_t x
       )
{
       /* This is a combination of tricks from "Hacker's Delight" with
        * some modifications, like a multiplication that rounds up to
        * drop the final adjustment stage.
        *
        * Do a partial reduction by digit sum to keep the value in the
        * range permitted for the mul/shift stage. There are several
        * possible and absolutely equivalent shift/mask combinations;
        * this one is ARM-friendly because of a mask that fits into 16
        * bit.
        */
       x = (x >> 15) + (x & UINT32_C(0x7FFF));
       /* Take reminder as (mod 8) by mul/shift. Since the multiplier
        * was calculated using ceil() instead of floor(), it skips the
        * value '7' properly.
        *    M <- ceil(ldexp(8/7, 29))
        */
       return (int)((x * UINT32_C(0x24924925)) >> 29);
}

int
i32mod7(
       int32_t x
       )
{
       /* We add (2**32 - 2**32 % 7), which is (2**32 - 4), to negative
        * numbers to map them into the postive range. Only the term '-4'
        * survives, obviously.
        */
       uint32_t ux = (uint32_t)x;
       return u32mod7((x < 0) ? (ux - 4u) : ux);
}

uint32_t
i32fmod(
       int32_t  x,
       uint32_t d
       )
{
       uint32_t ux = (uint32_t)x;
       uint32_t sf = UINT32_C(0) - (x < 0);
       ux = (sf ^ ux ) % d;
       return (d & sf) + (sf ^ ux);
}

/*
*---------------------------------------------------------------------
* Do a periodic extension of 'value' around 'pivot' with a period of
* 'cycle'.
*
* The result 'res' is a number that holds to the following properties:
*
*   1)  res MOD cycle == value MOD cycle
*   2)  pivot <= res < pivot + cycle
*       (replace </<= with >/>= for negative cycles)
*
* where 'MOD' denotes the modulo operator for FLOOR DIVISION, which
* is not the same as the '%' operator in C: C requires division to be
* a truncated division, where remainder and dividend have the same
* sign if the remainder is not zero, whereas floor division requires
* divider and modulus to have the same sign for a non-zero modulus.
*
* This function has some useful applications:
*
* + let Y be a calendar year and V a truncated 2-digit year: then
*      periodic_extend(Y-50, V, 100)
*   is the closest expansion of the truncated year with respect to
*   the full year, that is a 4-digit year with a difference of less
*   than 50 years to the year Y. ("century unfolding")
*
* + let T be a UN*X time stamp and V be seconds-of-day: then
*      perodic_extend(T-43200, V, 86400)
*   is a time stamp that has the same seconds-of-day as the input
*   value, with an absolute difference to T of <= 12hrs.  ("day
*   unfolding")
*
* + Wherever you have a truncated periodic value and a non-truncated
*   base value and you want to match them somehow...
*
* Basically, the function delivers 'pivot + (value - pivot) % cycle',
* but the implementation takes some pains to avoid internal signed
* integer overflows in the '(value - pivot) % cycle' part and adheres
* to the floor division convention.
*
* If 64bit scalars where available on all intended platforms, writing a
* version that uses 64 bit ops would be easy; writing a general
* division routine for 64bit ops on a platform that can only do
* 32/16bit divisions and is still performant is a bit more
* difficult. Since most usecases can be coded in a way that does only
* require the 32bit version a 64bit version is NOT provided here.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_periodic_extend(
       int32_t pivot,
       int32_t value,
       int32_t cycle
       )
{
       /* Implement a 4-quadrant modulus calculation by 2 2-quadrant
        * branches, one for positive and one for negative dividers.
        * Everything else can be handled by bit level logic and
        * conditional one's complement arithmetic.  By convention, we
        * assume
        *
        * x % b == 0  if  |b| < 2
        *
        * that is, we don't actually divide for cycles of -1,0,1 and
        * return the pivot value in that case.
        */
       uint32_t        uv = (uint32_t)value;
       uint32_t        up = (uint32_t)pivot;
       uint32_t        uc, sf;

       if (cycle > 1)
       {
               uc = (uint32_t)cycle;
               sf = UINT32_C(0) - (value < pivot);

               uv = sf ^ (uv - up);
               uv %= uc;
               pivot += (uc & sf) + (sf ^ uv);
       }
       else if (cycle < -1)
       {
               uc = ~(uint32_t)cycle + 1;
               sf = UINT32_C(0) - (value > pivot);

               uv = sf ^ (up - uv);
               uv %= uc;
               pivot -= (uc & sf) + (sf ^ uv);
       }
       return pivot;
}

/*---------------------------------------------------------------------
* Note to the casual reader
*
* In the next two functions you will find (or would have found...)
* the expression
*
*   res.Q_s -= 0x80000000;
*
* There was some ruckus about a possible programming error due to
* integer overflow and sign propagation.
*
* This assumption is based on a lack of understanding of the C
* standard. (Though this is admittedly not one of the most 'natural'
* aspects of the 'C' language and easily to get wrong.)
*
* see
*      http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf
*      "ISO/IEC 9899:201x Committee Draft — April 12, 2011"
*      6.4.4.1 Integer constants, clause 5
*
* why there is no sign extension/overflow problem here.
*
* But to ease the minds of the doubtful, I added back the 'u' qualifiers
* that somehow got lost over the last years.
*/


/*
*---------------------------------------------------------------------
* Convert a timestamp in NTP scale to a 64bit seconds value in the UN*X
* scale with proper epoch unfolding around a given pivot or the current
* system time. This function happily accepts negative pivot values as
* timestamps before 1970-01-01, so be aware of possible trouble on
* platforms with 32bit 'time_t'!
*
* This is also a periodic extension, but since the cycle is 2^32 and
* the shift is 2^31, we can do some *very* fast math without explicit
* divisions.
*---------------------------------------------------------------------
*/
vint64
ntpcal_ntp_to_time(
       uint32_t        ntp,
       const time_t *  pivot
       )
{
       vint64 res;

#   if defined(HAVE_INT64)

       res.q_s = (pivot != NULL)
                     ? *pivot
                     : now();
       res.Q_s -= 0x80000000u;         /* unshift of half range */
       ntp     -= (uint32_t)JAN_1970;  /* warp into UN*X domain */
       ntp     -= res.D_s.lo;          /* cycle difference      */
       res.Q_s += (uint64_t)ntp;       /* get expanded time     */

#   else /* no 64bit scalars */

       time_t tmp;

       tmp = (pivot != NULL)
                 ? *pivot
                 : now();
       res = time_to_vint64(&tmp);
       M_SUB(res.D_s.hi, res.D_s.lo, 0, 0x80000000u);
       ntp -= (uint32_t)JAN_1970;      /* warp into UN*X domain */
       ntp -= res.D_s.lo;              /* cycle difference      */
       M_ADD(res.D_s.hi, res.D_s.lo, 0, ntp);

#   endif /* no 64bit scalars */

       return res;
}

/*
*---------------------------------------------------------------------
* Convert a timestamp in NTP scale to a 64bit seconds value in the NTP
* scale with proper epoch unfolding around a given pivot or the current
* system time.
*
* Note: The pivot must be given in the UN*X time domain!
*
* This is also a periodic extension, but since the cycle is 2^32 and
* the shift is 2^31, we can do some *very* fast math without explicit
* divisions.
*---------------------------------------------------------------------
*/
vint64
ntpcal_ntp_to_ntp(
       uint32_t      ntp,
       const time_t *pivot
       )
{
       vint64 res;

#   if defined(HAVE_INT64)

       res.q_s = (pivot)
                     ? *pivot
                     : now();
       res.Q_s -= 0x80000000u;         /* unshift of half range */
       res.Q_s += (uint32_t)JAN_1970;  /* warp into NTP domain  */
       ntp     -= res.D_s.lo;          /* cycle difference      */
       res.Q_s += (uint64_t)ntp;       /* get expanded time     */

#   else /* no 64bit scalars */

       time_t tmp;

       tmp = (pivot)
                 ? *pivot
                 : now();
       res = time_to_vint64(&tmp);
       M_SUB(res.D_s.hi, res.D_s.lo, 0, 0x80000000u);
       M_ADD(res.D_s.hi, res.D_s.lo, 0, (uint32_t)JAN_1970);/*into NTP */
       ntp -= res.D_s.lo;              /* cycle difference      */
       M_ADD(res.D_s.hi, res.D_s.lo, 0, ntp);

#   endif /* no 64bit scalars */

       return res;
}


/*
* ====================================================================
*
* Splitting values to composite entities
*
* ====================================================================
*/

/*
*---------------------------------------------------------------------
* Split a 64bit seconds value into elapsed days in 'res.hi' and
* elapsed seconds since midnight in 'res.lo' using explicit floor
* division. This function happily accepts negative time values as
* timestamps before the respective epoch start.
*---------------------------------------------------------------------
*/
ntpcal_split
ntpcal_daysplit(
       const vint64 *ts
       )
{
       ntpcal_split res;
       uint32_t Q, R;

#   if defined(HAVE_64BITREGS)

       /* Assume we have 64bit registers an can do a divison by
        * constant reasonably fast using the one's complement trick..
        */
       uint64_t sf64 = (uint64_t)-(ts->q_s < 0);
       Q = (uint32_t)(sf64 ^ ((sf64 ^ ts->Q_s) / SECSPERDAY));
       R = (uint32_t)(ts->Q_s - Q * SECSPERDAY);

#   elif defined(UINT64_MAX) && !defined(__arm__)

       /* We rely on the compiler to do efficient 64bit divisions as
        * good as possible. Which might or might not be true. At least
        * for ARM CPUs, the sum-by-digit code in the next section is
        * faster for many compilers. (This might change over time, but
        * the 64bit-by-32bit division will never outperform the exact
        * division by a substantial factor....)
        */
       if (ts->q_s < 0)
               Q = ~(uint32_t)(~ts->Q_s / SECSPERDAY);
       else
               Q =  (uint32_t)( ts->Q_s / SECSPERDAY);
       R = ts->D_s.lo - Q * SECSPERDAY;

#   else

       /* We don't have 64bit regs. That hurts a bit.
        *
        * Here we use a mean trick to get away with just one explicit
        * modulo operation and pure 32bit ops.
        *
        * Remember: 86400 <--> 128 * 675
        *
        * So we discard the lowest 7 bit and do an exact division by
        * 675, modulo 2**32.
        *
        * First we shift out the lower 7 bits.
        *
        * Then we use a digit-wise pseudo-reduction, where a 'digit' is
        * actually a 16-bit group. This is followed by a full reduction
        * with a 'true' division step. This yields the modulus of the
        * full 64bit value. The sign bit gets some extra treatment.
        *
        * Then we decrement the lower limb by that modulus, so it is
        * exactly divisible by 675. [*]
        *
        * Then we multiply with the modular inverse of 675 (mod 2**32)
        * and voila, we have the result.
        *
        * Special Thanks to Henry S. Warren and his "Hacker's delight"
        * for giving that idea.
        *
        * (Note[*]: that's not the full truth. We would have to
        * subtract the modulus from the full 64 bit number to get a
        * number that is divisible by 675. But since we use the
        * multiplicative inverse (mod 2**32) there's no reason to carry
        * the subtraction into the upper bits!)
        */
       uint32_t al = ts->D_s.lo;
       uint32_t ah = ts->D_s.hi;

       /* shift out the lower 7 bits, smash sign bit */
       al = (al >> 7) | (ah << 25);
       ah = (ah >> 7) & 0x00FFFFFFu;

       R  = (ts->d_s.hi < 0) ? 239 : 0;/* sign bit value */
       R += (al & 0xFFFF);
       R += (al >> 16   ) * 61u;       /* 2**16 % 675 */
       R += (ah & 0xFFFF) * 346u;      /* 2**32 % 675 */
       R += (ah >> 16   ) * 181u;      /* 2**48 % 675 */
       R %= 675u;                      /* final reduction */
       Q  = (al - R) * 0x2D21C10Bu;    /* modinv(675, 2**32) */
       R  = (R << 7) | (ts->d_s.lo & 0x07F);

#   endif

       res.hi = uint32_2cpl_to_int32(Q);
       res.lo = R;

       return res;
}

/*
*---------------------------------------------------------------------
* Split a 64bit seconds value into elapsed weeks in 'res.hi' and
* elapsed seconds since week start in 'res.lo' using explicit floor
* division. This function happily accepts negative time values as
* timestamps before the respective epoch start.
*---------------------------------------------------------------------
*/
ntpcal_split
ntpcal_weeksplit(
       const vint64 *ts
       )
{
       ntpcal_split res;
       uint32_t Q, R;

       /* This is a very close relative to the day split function; for
        * details, see there!
        */

#   if defined(HAVE_64BITREGS)

       uint64_t sf64 = (uint64_t)-(ts->q_s < 0);
       Q = (uint32_t)(sf64 ^ ((sf64 ^ ts->Q_s) / SECSPERWEEK));
       R = (uint32_t)(ts->Q_s - Q * SECSPERWEEK);

#   elif defined(UINT64_MAX) && !defined(__arm__)

       if (ts->q_s < 0)
               Q = ~(uint32_t)(~ts->Q_s / SECSPERWEEK);
       else
               Q =  (uint32_t)( ts->Q_s / SECSPERWEEK);
       R = ts->D_s.lo - Q * SECSPERWEEK;

#   else

       /* Remember: 7*86400 <--> 604800 <--> 128 * 4725 */
       uint32_t al = ts->D_s.lo;
       uint32_t ah = ts->D_s.hi;

       al = (al >> 7) | (ah << 25);
       ah = (ah >> 7) & 0x00FFFFFF;

       R  = (ts->d_s.hi < 0) ? 2264 : 0;/* sign bit value */
       R += (al & 0xFFFF);
       R += (al >> 16   ) * 4111u;     /* 2**16 % 4725 */
       R += (ah & 0xFFFF) * 3721u;     /* 2**32 % 4725 */
       R += (ah >> 16   ) * 2206u;     /* 2**48 % 4725 */
       R %= 4725u;                     /* final reduction */
       Q  = (al - R) * 0x98BBADDDu;    /* modinv(4725, 2**32) */
       R  = (R << 7) | (ts->d_s.lo & 0x07F);

#   endif

       res.hi = uint32_2cpl_to_int32(Q);
       res.lo = R;

       return res;
}

/*
*---------------------------------------------------------------------
* Split a 32bit seconds value into h/m/s and excessive days.  This
* function happily accepts negative time values as timestamps before
* midnight.
*---------------------------------------------------------------------
*/
static int32_t
priv_timesplit(
       int32_t split[3],
       int32_t ts
       )
{
       /* Do 3 chained floor divisions by positive constants, using the
        * one's complement trick and factoring out the intermediate XOR
        * ops to reduce the number of operations.
        */
       uint32_t us, um, uh, ud, sf32;

       sf32 = int32_sflag(ts);

       us = (uint32_t)ts;
       um = (sf32 ^ us) / SECSPERMIN;
       uh = um / MINSPERHR;
       ud = uh / HRSPERDAY;

       um ^= sf32;
       uh ^= sf32;
       ud ^= sf32;

       split[0] = (int32_t)(uh - ud * HRSPERDAY );
       split[1] = (int32_t)(um - uh * MINSPERHR );
       split[2] = (int32_t)(us - um * SECSPERMIN);

       return uint32_2cpl_to_int32(ud);
}

/*
*---------------------------------------------------------------------
* Given the number of elapsed days in the calendar era, split this
* number into the number of elapsed years in 'res.hi' and the number
* of elapsed days of that year in 'res.lo'.
*
* if 'isleapyear' is not NULL, it will receive an integer that is 0 for
* regular years and a non-zero value for leap years.
*---------------------------------------------------------------------
*/
ntpcal_split
ntpcal_split_eradays(
       int32_t days,
       int  *isleapyear
       )
{
       /* Use the fast cycle split algorithm here, to calculate the
        * centuries and years in a century with one division each. This
        * reduces the number of division operations to two, but is
        * susceptible to internal range overflow. We take some extra
        * steps to avoid the gap.
        */
       ntpcal_split res;
       int32_t  n100, n001; /* calendar year cycles */
       uint32_t uday, Q;

       /* split off centuries first
        *
        * We want to execute '(days * 4 + 3) /% 146097' under floor
        * division rules in the first step. Well, actually we want to
        * calculate 'floor((days + 0.75) / 36524.25)', but we want to
        * do it in scaled integer calculation.
        */
#   if defined(HAVE_64BITREGS)

       /* not too complicated with an intermediate 64bit value */
       uint64_t        ud64, sf64;
       ud64 = ((uint64_t)days << 2) | 3u;
       sf64 = (uint64_t)-(days < 0);
       Q    = (uint32_t)(sf64 ^ ((sf64 ^ ud64) / GREGORIAN_CYCLE_DAYS));
       uday = (uint32_t)(ud64 - Q * GREGORIAN_CYCLE_DAYS);
       n100 = uint32_2cpl_to_int32(Q);

#   else

       /* '4*days+3' suffers from range overflow when going to the
        * limits. We solve this by doing an exact division (mod 2^32)
        * after caclulating the remainder first.
        *
        * We start with a partial reduction by digit sums, extracting
        * the upper bits from the original value before they get lost
        * by scaling, and do one full division step to get the true
        * remainder.  Then a final multiplication with the
        * multiplicative inverse of 146097 (mod 2^32) gives us the full
        * quotient.
        *
        * (-2^33) % 146097     --> 130717    : the sign bit value
        * ( 2^20) % 146097     --> 25897     : the upper digit value
        * modinv(146097, 2^32) --> 660721233 : the inverse
        */
       uint32_t ux = ((uint32_t)days << 2) | 3;
       uday  = (days < 0) ? 130717u : 0u;          /* sign dgt */
       uday += ((days >> 18) & 0x01FFFu) * 25897u; /* hi dgt (src!) */
       uday += (ux & 0xFFFFFu);                    /* lo dgt */
       uday %= GREGORIAN_CYCLE_DAYS;               /* full reduction */
       Q     = (ux  - uday) * 660721233u;          /* exact div */
       n100  = uint32_2cpl_to_int32(Q);

#   endif

       /* Split off years in century -- days >= 0 here, and we're far
        * away from integer overflow trouble now. */
       uday |= 3;
       n001  = uday / GREGORIAN_NORMAL_LEAP_CYCLE_DAYS;
       uday -= n001 * GREGORIAN_NORMAL_LEAP_CYCLE_DAYS;

       /* Assemble the year and day in year */
       res.hi = n100 * 100 + n001;
       res.lo = uday / 4u;

       /* Possibly set the leap year flag */
       if (isleapyear) {
               uint32_t tc = (uint32_t)n100 + 1;
               uint32_t ty = (uint32_t)n001 + 1;
               *isleapyear = !(ty & 3)
                   && ((ty != 100) || !(tc & 3));
       }
       return res;
}

/*
*---------------------------------------------------------------------
* Given a number of elapsed days in a year and a leap year indicator,
* split the number of elapsed days into the number of elapsed months in
* 'res.hi' and the number of elapsed days of that month in 'res.lo'.
*
* This function will fail and return {-1,-1} if the number of elapsed
* days is not in the valid range!
*---------------------------------------------------------------------
*/
ntpcal_split
ntpcal_split_yeardays(
       int32_t eyd,
       int     isleap
       )
{
       /* Use the unshifted-year, February-with-30-days approach here.
        * Fractional interpolations are used in both directions, with
        * the smallest power-of-two divider to avoid any true division.
        */
       ntpcal_split    res = {-1, -1};

       /* convert 'isleap' to number of defective days */
       isleap = 1 + !isleap;
       /* adjust for February of 30 nominal days */
       if (eyd >= 61 - isleap)
               eyd += isleap;
       /* if in range, convert to months and days in month */
       if (eyd >= 0 && eyd < 367) {
               res.hi = (eyd * 67 + 32) >> 11;
               res.lo = eyd - ((489 * res.hi + 8) >> 4);
       }

       return res;
}

/*
*---------------------------------------------------------------------
* Convert a RD into the date part of a 'struct calendar'.
*---------------------------------------------------------------------
*/
int
ntpcal_rd_to_date(
       struct calendar *jd,
       int32_t          rd
       )
{
       ntpcal_split split;
       int          leapy;
       u_int        ymask;

       /* Get day-of-week first. It's simply the RD (mod 7)... */
       jd->weekday = i32mod7(rd);

       split = ntpcal_split_eradays(rd - 1, &leapy);
       /* Get year and day-of-year, with overflow check. If any of the
        * upper 16 bits is set after shifting to unity-based years, we
        * will have an overflow when converting to an unsigned 16bit
        * year. Shifting to the right is OK here, since it does not
        * matter if the shift is logic or arithmetic.
        */
       split.hi += 1;
       ymask = 0u - ((split.hi >> 16) == 0);
       jd->year = (uint16_t)(split.hi & ymask);
       jd->yearday = (uint16_t)split.lo + 1;

       /* convert to month and mday */
       split = ntpcal_split_yeardays(split.lo, leapy);
       jd->month    = (uint8_t)split.hi + 1;
       jd->monthday = (uint8_t)split.lo + 1;

       return ymask ? leapy : -1;
}

/*
*---------------------------------------------------------------------
* Convert a RD into the date part of a 'struct tm'.
*---------------------------------------------------------------------
*/
int
ntpcal_rd_to_tm(
       struct tm  *utm,
       int32_t     rd
       )
{
       ntpcal_split split;
       int          leapy;

       /* get day-of-week first */
       utm->tm_wday = i32mod7(rd);

       /* get year and day-of-year */
       split = ntpcal_split_eradays(rd - 1, &leapy);
       utm->tm_year = split.hi - 1899;
       utm->tm_yday = split.lo;        /* 0-based */

       /* convert to month and mday */
       split = ntpcal_split_yeardays(split.lo, leapy);
       utm->tm_mon  = split.hi;        /* 0-based */
       utm->tm_mday = split.lo + 1;    /* 1-based */

       return leapy;
}

/*
*---------------------------------------------------------------------
* Take a value of seconds since midnight and split it into hhmmss in a
* 'struct calendar'.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_daysec_to_date(
       struct calendar *jd,
       int32_t         sec
       )
{
       int32_t days;
       int   ts[3];

       days = priv_timesplit(ts, sec);
       jd->hour   = (uint8_t)ts[0];
       jd->minute = (uint8_t)ts[1];
       jd->second = (uint8_t)ts[2];

       return days;
}

/*
*---------------------------------------------------------------------
* Take a value of seconds since midnight and split it into hhmmss in a
* 'struct tm'.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_daysec_to_tm(
       struct tm *utm,
       int32_t    sec
       )
{
       int32_t days;
       int32_t ts[3];

       days = priv_timesplit(ts, sec);
       utm->tm_hour = ts[0];
       utm->tm_min  = ts[1];
       utm->tm_sec  = ts[2];

       return days;
}

/*
*---------------------------------------------------------------------
* take a split representation for day/second-of-day and day offset
* and convert it to a 'struct calendar'. The seconds will be normalised
* into the range of a day, and the day will be adjusted accordingly.
*
* returns >0 if the result is in a leap year, 0 if in a regular
* year and <0 if the result did not fit into the calendar struct.
*---------------------------------------------------------------------
*/
int
ntpcal_daysplit_to_date(
       struct calendar    *jd,
       const ntpcal_split *ds,
       int32_t             dof
       )
{
       dof += ntpcal_daysec_to_date(jd, ds->lo);
       return ntpcal_rd_to_date(jd, ds->hi + dof);
}

/*
*---------------------------------------------------------------------
* take a split representation for day/second-of-day and day offset
* and convert it to a 'struct tm'. The seconds will be normalised
* into the range of a day, and the day will be adjusted accordingly.
*
* returns 1 if the result is in a leap year and zero if in a regular
* year.
*---------------------------------------------------------------------
*/
int
ntpcal_daysplit_to_tm(
       struct tm          *utm,
       const ntpcal_split *ds ,
       int32_t             dof
       )
{
       dof += ntpcal_daysec_to_tm(utm, ds->lo);

       return ntpcal_rd_to_tm(utm, ds->hi + dof);
}

/*
*---------------------------------------------------------------------
* Take a UN*X time and convert to a calendar structure.
*---------------------------------------------------------------------
*/
int
ntpcal_time_to_date(
       struct calendar *jd,
       const vint64    *ts
       )
{
       ntpcal_split ds;

       ds = ntpcal_daysplit(ts);
       ds.hi += ntpcal_daysec_to_date(jd, ds.lo);
       ds.hi += DAY_UNIX_STARTS;

       return ntpcal_rd_to_date(jd, ds.hi);
}


/*
* ====================================================================
*
* merging composite entities
*
* ====================================================================
*/

#if !defined(HAVE_INT64)
/* multiplication helper. Seconds in days and weeks are multiples of 128,
* and without that factor fit well into 16 bit. So a multiplication
* of 32bit by 16bit and some shifting can be used on pure 32bit machines
* with compilers that do not support 64bit integers.
*
* Calculate ( hi * mul * 128 ) + lo
*/
static vint64
_dwjoin(
       uint16_t        mul,
       int32_t         hi,
       int32_t         lo
       )
{
       vint64          res;
       uint32_t        p1, p2, sf;

       /* get sign flag and absolute value of 'hi' in p1 */
       sf = (uint32_t)-(hi < 0);
       p1 = ((uint32_t)hi + sf) ^ sf;

       /* assemble major units: res <- |hi| * mul */
       res.D_s.lo = (p1 & 0xFFFF) * mul;
       res.D_s.hi = 0;
       p1 = (p1 >> 16) * mul;
       p2 = p1 >> 16;
       p1 = p1 << 16;
       M_ADD(res.D_s.hi, res.D_s.lo, p2, p1);

       /* mul by 128, using shift: res <-- res << 7 */
       res.D_s.hi = (res.D_s.hi << 7) | (res.D_s.lo >> 25);
       res.D_s.lo = (res.D_s.lo << 7);

       /* fix up sign: res <-- (res + [sf|sf]) ^ [sf|sf] */
       M_ADD(res.D_s.hi, res.D_s.lo, sf, sf);
       res.D_s.lo ^= sf;
       res.D_s.hi ^= sf;

       /* properly add seconds: res <-- res + [sx(lo)|lo] */
       p2 = (uint32_t)-(lo < 0);
       p1 = (uint32_t)lo;
       M_ADD(res.D_s.hi, res.D_s.lo, p2, p1);
       return res;
}
#endif

/*
*---------------------------------------------------------------------
* Merge a number of days and a number of seconds into seconds,
* expressed in 64 bits to avoid overflow.
*---------------------------------------------------------------------
*/
vint64
ntpcal_dayjoin(
       int32_t days,
       int32_t secs
       )
{
       vint64 res;

#   if defined(HAVE_INT64)

       res.q_s  = days;
       res.q_s *= SECSPERDAY;
       res.q_s += secs;

#   else

       res = _dwjoin(675, days, secs);

#   endif

       return res;
}

/*
*---------------------------------------------------------------------
* Merge a number of weeks and a number of seconds into seconds,
* expressed in 64 bits to avoid overflow.
*---------------------------------------------------------------------
*/
vint64
ntpcal_weekjoin(
       int32_t week,
       int32_t secs
       )
{
       vint64 res;

#   if defined(HAVE_INT64)

       res.q_s  = week;
       res.q_s *= SECSPERWEEK;
       res.q_s += secs;

#   else

       res = _dwjoin(4725, week, secs);

#   endif

       return res;
}

/*
*---------------------------------------------------------------------
* get leap years since epoch in elapsed years
*---------------------------------------------------------------------
*/
int32_t
ntpcal_leapyears_in_years(
       int32_t years
       )
{
       /* We use the in-out-in algorithm here, using the one's
        * complement division trick for negative numbers. The chained
        * division sequence by 4/25/4 gives the compiler the chance to
        * get away with only one true division and doing shifts otherwise.
        */

       uint32_t sf32, sum, uyear;

       sf32  = int32_sflag(years);
       uyear = (uint32_t)years;
       uyear ^= sf32;

       sum  = (uyear /=  4u);  /*   4yr rule --> IN  */
       sum -= (uyear /= 25u);  /* 100yr rule --> OUT */
       sum += (uyear /=  4u);  /* 400yr rule --> IN  */

       /* Thanks to the alternation of IN/OUT/IN we can do the sum
        * directly and have a single one's complement operation
        * here. (Only if the years are negative, of course.) Otherwise
        * the one's complement would have to be done when
        * adding/subtracting the terms.
        */
       return uint32_2cpl_to_int32(sf32 ^ sum);
}

/*
*---------------------------------------------------------------------
* Convert elapsed years in Era into elapsed days in Era.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_days_in_years(
       int32_t years
       )
{
       return years * DAYSPERYEAR + ntpcal_leapyears_in_years(years);
}

/*
*---------------------------------------------------------------------
* Convert a number of elapsed month in a year into elapsed days in year.
*
* The month will be normalized, and 'res.hi' will contain the
* excessive years that must be considered when converting the years,
* while 'res.lo' will contain the number of elapsed days since start
* of the year.
*
* This code uses the shifted-month-approach to convert month to days,
* because then there is no need to have explicit leap year
* information.  The slight disadvantage is that for most month values
* the result is a negative value, and the year excess is one; the
* conversion is then simply based on the start of the following year.
*---------------------------------------------------------------------
*/
ntpcal_split
ntpcal_days_in_months(
       int32_t m
       )
{
       ntpcal_split res;

       /* Add ten months with proper year adjustment. */
       if (m < 2) {
           res.lo  = m + 10;
           res.hi  = 0;
       } else {
           res.lo  = m - 2;
           res.hi  = 1;
       }

       /* Possibly normalise by floor division. This does not hapen for
        * input in normal range. */
       if (res.lo < 0 || res.lo >= 12) {
               uint32_t mu, Q, sf32;
               sf32 = int32_sflag(res.lo);
               mu   = (uint32_t)res.lo;
               Q    = sf32 ^ ((sf32 ^ mu) / 12u);

               res.hi += uint32_2cpl_to_int32(Q);
               res.lo  = mu - Q * 12u;
       }

       /* Get cummulated days in year with unshift. Use the fractional
        * interpolation with smallest possible power of two in the
        * divider.
        */
       res.lo = ((res.lo * 979 + 16) >> 5) - 306;

       return res;
}

/*
*---------------------------------------------------------------------
* Convert ELAPSED years/months/days of gregorian calendar to elapsed
* days in Gregorian epoch.
*
* If you want to convert years and days-of-year, just give a month of
* zero.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_edate_to_eradays(
       int32_t years,
       int32_t mons,
       int32_t mdays
       )
{
       ntpcal_split tmp;
       int32_t      res;

       if (mons) {
               tmp = ntpcal_days_in_months(mons);
               res = ntpcal_days_in_years(years + tmp.hi) + tmp.lo;
       } else
               res = ntpcal_days_in_years(years);
       res += mdays;

       return res;
}

/*
*---------------------------------------------------------------------
* Convert ELAPSED years/months/days of gregorian calendar to elapsed
* days in year.
*
* Note: This will give the true difference to the start of the given
* year, even if months & days are off-scale.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_edate_to_yeardays(
       int32_t years,
       int32_t mons,
       int32_t mdays
       )
{
       ntpcal_split tmp;

       if (0 <= mons && mons < 12) {
               if (mons >= 2)
                       mdays -= 2 - is_leapyear(years+1);
               mdays += (489 * mons + 8) >> 4;
       } else {
               tmp = ntpcal_days_in_months(mons);
               mdays += tmp.lo
                      + ntpcal_days_in_years(years + tmp.hi)
                      - ntpcal_days_in_years(years);
       }

       return mdays;
}

/*
*---------------------------------------------------------------------
* Convert elapsed days and the hour/minute/second information into
* total seconds.
*
* If 'isvalid' is not NULL, do a range check on the time specification
* and tell if the time input is in the normal range, permitting for a
* single leapsecond.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_etime_to_seconds(
       int32_t hours,
       int32_t minutes,
       int32_t seconds
       )
{
       int32_t res;

       res = (hours * MINSPERHR + minutes) * SECSPERMIN + seconds;

       return res;
}

/*
*---------------------------------------------------------------------
* Convert the date part of a 'struct tm' (that is, year, month,
* day-of-month) into the RD of that day.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_tm_to_rd(
       const struct tm *utm
       )
{
       return ntpcal_edate_to_eradays(utm->tm_year + 1899,
                                      utm->tm_mon,
                                      utm->tm_mday - 1) + 1;
}

/*
*---------------------------------------------------------------------
* Convert the date part of a 'struct calendar' (that is, year, month,
* day-of-month) into the RD of that day.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_date_to_rd(
       const struct calendar *jd
       )
{
       return ntpcal_edate_to_eradays((int32_t)jd->year - 1,
                                      (int32_t)jd->month - 1,
                                      (int32_t)jd->monthday - 1) + 1;
}

/*
*---------------------------------------------------------------------
* convert a year number to rata die of year start
*---------------------------------------------------------------------
*/
int32_t
ntpcal_year_to_ystart(
       int32_t year
       )
{
       return ntpcal_days_in_years(year - 1) + 1;
}

/*
*---------------------------------------------------------------------
* For a given RD, get the RD of the associated year start,
* that is, the RD of the last January,1st on or before that day.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_rd_to_ystart(
       int32_t rd
       )
{
       /*
        * Rather simple exercise: split the day number into elapsed
        * years and elapsed days, then remove the elapsed days from the
        * input value. Nice'n sweet...
        */
       return rd - ntpcal_split_eradays(rd - 1, NULL).lo;
}

/*
*---------------------------------------------------------------------
* For a given RD, get the RD of the associated month start.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_rd_to_mstart(
       int32_t rd
       )
{
       ntpcal_split split;
       int          leaps;

       split = ntpcal_split_eradays(rd - 1, &leaps);
       split = ntpcal_split_yeardays(split.lo, leaps);

       return rd - split.lo;
}

/*
*---------------------------------------------------------------------
* take a 'struct calendar' and get the seconds-of-day from it.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_date_to_daysec(
       const struct calendar *jd
       )
{
       return ntpcal_etime_to_seconds(jd->hour, jd->minute,
                                      jd->second);
}

/*
*---------------------------------------------------------------------
* take a 'struct tm' and get the seconds-of-day from it.
*---------------------------------------------------------------------
*/
int32_t
ntpcal_tm_to_daysec(
       const struct tm *utm
       )
{
       return ntpcal_etime_to_seconds(utm->tm_hour, utm->tm_min,
                                      utm->tm_sec);
}

/*
*---------------------------------------------------------------------
* take a 'struct calendar' and convert it to a 'time_t'
*---------------------------------------------------------------------
*/
time_t
ntpcal_date_to_time(
       const struct calendar *jd
       )
{
       vint64  join;
       int32_t days, secs;

       days = ntpcal_date_to_rd(jd) - DAY_UNIX_STARTS;
       secs = ntpcal_date_to_daysec(jd);
       join = ntpcal_dayjoin(days, secs);

       return vint64_to_time(&join);
}


/*
* ====================================================================
*
* extended and unchecked variants of caljulian/caltontp
*
* ====================================================================
*/
int
ntpcal_ntp64_to_date(
       struct calendar *jd,
       const vint64    *ntp
       )
{
       ntpcal_split ds;

       ds = ntpcal_daysplit(ntp);
       ds.hi += ntpcal_daysec_to_date(jd, ds.lo);

       return ntpcal_rd_to_date(jd, ds.hi + DAY_NTP_STARTS);
}

int
ntpcal_ntp_to_date(
       struct calendar *jd,
       uint32_t         ntp,
       const time_t    *piv
       )
{
       vint64  ntp64;

       /*
        * Unfold ntp time around current time into NTP domain. Split
        * into days and seconds, shift days into CE domain and
        * process the parts.
        */
       ntp64 = ntpcal_ntp_to_ntp(ntp, piv);
       return ntpcal_ntp64_to_date(jd, &ntp64);
}


vint64
ntpcal_date_to_ntp64(
       const struct calendar *jd
       )
{
       /*
        * Convert date to NTP. Ignore yearday, use d/m/y only.
        */
       return ntpcal_dayjoin(ntpcal_date_to_rd(jd) - DAY_NTP_STARTS,
                             ntpcal_date_to_daysec(jd));
}


uint32_t
ntpcal_date_to_ntp(
       const struct calendar *jd
       )
{
       /*
        * Get lower half of 64bit NTP timestamp from date/time.
        */
       return ntpcal_date_to_ntp64(jd).d_s.lo;
}



/*
* ====================================================================
*
* day-of-week calculations
*
* ====================================================================
*/
/*
* Given a RataDie and a day-of-week, calculate a RDN that is reater-than,
* greater-or equal, closest, less-or-equal or less-than the given RDN
* and denotes the given day-of-week
*/
int32_t
ntpcal_weekday_gt(
       int32_t rdn,
       int32_t dow
       )
{
       return ntpcal_periodic_extend(rdn+1, dow, 7);
}

int32_t
ntpcal_weekday_ge(
       int32_t rdn,
       int32_t dow
       )
{
       return ntpcal_periodic_extend(rdn, dow, 7);
}

int32_t
ntpcal_weekday_close(
       int32_t rdn,
       int32_t dow
       )
{
       return ntpcal_periodic_extend(rdn-3, dow, 7);
}

int32_t
ntpcal_weekday_le(
       int32_t rdn,
       int32_t dow
       )
{
       return ntpcal_periodic_extend(rdn, dow, -7);
}

int32_t
ntpcal_weekday_lt(
       int32_t rdn,
       int32_t dow
       )
{
       return ntpcal_periodic_extend(rdn-1, dow, -7);
}

/*
* ====================================================================
*
* ISO week-calendar conversions
*
* The ISO8601 calendar defines a calendar of years, weeks and weekdays.
* It is related to the Gregorian calendar, and a ISO year starts at the
* Monday closest to Jan,1st of the corresponding Gregorian year.  A ISO
* calendar year has always 52 or 53 weeks, and like the Grogrian
* calendar the ISO8601 calendar repeats itself every 400 years, or
* 146097 days, or 20871 weeks.
*
* While it is possible to write ISO calendar functions based on the
* Gregorian calendar functions, the following implementation takes a
* different approach, based directly on years and weeks.
*
* Analysis of the tabulated data shows that it is not possible to
* interpolate from years to weeks over a full 400 year range; cyclic
* shifts over 400 years do not provide a solution here. But it *is*
* possible to interpolate over every single century of the 400-year
* cycle. (The centennial leap year rule seems to be the culprit here.)
*
* It can be shown that a conversion from years to weeks can be done
* using a linear transformation of the form
*
*   w = floor( y * a + b )
*
* where the slope a must hold to
*
*  52.1780821918 <= a < 52.1791044776
*
* and b must be chosen according to the selected slope and the number
* of the century in a 400-year period.
*
* The inverse calculation can also be done in this way. Careful scaling
* provides an unlimited set of integer coefficients a,k,b that enable
* us to write the calulation in the form
*
*   w = (y * a  + b ) / k
*   y = (w * a' + b') / k'
*
* In this implementation the values of k and k' are chosen to be the
* smallest possible powers of two, so the division can be implemented
* as shifts if the optimiser chooses to do so.
*
* ====================================================================
*/

/*
* Given a number of elapsed (ISO-)years since the begin of the
* christian era, return the number of elapsed weeks corresponding to
* the number of years.
*/
int32_t
isocal_weeks_in_years(
       int32_t years
       )
{
       /*
        * use: w = (y * 53431 + b[c]) / 1024 as interpolation
        */
       static const uint16_t bctab[4] = { 157, 449, 597, 889 };

       int32_t  cs, cw;
       uint32_t cc, ci, yu, sf32;

       sf32 = int32_sflag(years);
       yu   = (uint32_t)years;

       /* split off centuries, using floor division */
       cc  = sf32 ^ ((sf32 ^ yu) / 100u);
       yu -= cc * 100u;

       /* calculate century cycles shift and cycle index:
        * Assuming a century is 5217 weeks, we have to add a cycle
        * shift that is 3 for every 4 centuries, because 3 of the four
        * centuries have 5218 weeks. So '(cc*3 + 1) / 4' is the actual
        * correction, and the second century is the defective one.
        *
        * Needs floor division by 4, which is done with masking and
        * shifting.
        */
       ci = cc * 3u + 1;
       cs = uint32_2cpl_to_int32(sf32 ^ ((sf32 ^ ci) >> 2));
       ci = ci & 3u;

       /* Get weeks in century. Can use plain division here as all ops
        * are >= 0,  and let the compiler sort out the possible
        * optimisations.
        */
       cw = (yu * 53431u + bctab[ci]) / 1024u;

       return uint32_2cpl_to_int32(cc) * 5217 + cs + cw;
}

/*
* Given a number of elapsed weeks since the begin of the christian
* era, split this number into the number of elapsed years in res.hi
* and the excessive number of weeks in res.lo. (That is, res.lo is
* the number of elapsed weeks in the remaining partial year.)
*/
ntpcal_split
isocal_split_eraweeks(
       int32_t weeks
       )
{
       /*
        * use: y = (w * 157 + b[c]) / 8192 as interpolation
        */

       static const uint16_t bctab[4] = { 85, 130, 17, 62 };

       ntpcal_split res;
       int32_t  cc, ci;
       uint32_t sw, cy, Q;

       /* Use two fast cycle-split divisions again. Herew e want to
        * execute '(weeks * 4 + 2) /% 20871' under floor division rules
        * in the first step.
        *
        * This is of course (again) susceptible to internal overflow if
        * coded directly in 32bit. And again we use 64bit division on
        * a 64bit target and exact division after calculating the
        * remainder first on a 32bit target. With the smaller divider,
        * that's even a bit neater.
        */
#   if defined(HAVE_64BITREGS)

       /* Full floor division with 64bit values. */
       uint64_t sf64, sw64;
       sf64 = (uint64_t)-(weeks < 0);
       sw64 = ((uint64_t)weeks << 2) | 2u;
       Q    = (uint32_t)(sf64 ^ ((sf64 ^ sw64) / GREGORIAN_CYCLE_WEEKS));
       sw   = (uint32_t)(sw64 - Q * GREGORIAN_CYCLE_WEEKS);

#   else

       /* Exact division after calculating the remainder via partial
        * reduction by digit sum.
        * (-2^33) % 20871     --> 5491      : the sign bit value
        * ( 2^20) % 20871     --> 5026      : the upper digit value
        * modinv(20871, 2^32) --> 330081335 : the inverse
        */
       uint32_t ux = ((uint32_t)weeks << 2) | 2;
       sw  = (weeks < 0) ? 5491u : 0u;           /* sign dgt */
       sw += ((weeks >> 18) & 0x01FFFu) * 5026u; /* hi dgt (src!) */
       sw += (ux & 0xFFFFFu);                    /* lo dgt */
       sw %= GREGORIAN_CYCLE_WEEKS;              /* full reduction */
       Q   = (ux  - sw) * 330081335u;            /* exact div */

#   endif

       ci  = Q & 3u;
       cc  = uint32_2cpl_to_int32(Q);

       /* Split off years; sw >= 0 here! The scaled weeks in the years
        * are scaled up by 157 afterwards.
        */
       sw  = (sw / 4u) * 157u + bctab[ci];
       cy  = sw / 8192u;       /* sw >> 13 , let the compiler sort it out */
       sw  = sw % 8192u;       /* sw & 8191, let the compiler sort it out */

       /* assemble elapsed years and downscale the elapsed weeks in
        * the year.
        */
       res.hi = 100*cc + cy;
       res.lo = sw / 157u;

       return res;
}

/*
* Given a second in the NTP time scale and a pivot, expand the NTP
* time stamp around the pivot and convert into an ISO calendar time
* stamp.
*/
int
isocal_ntp64_to_date(
       struct isodate *id,
       const vint64   *ntp
       )
{
       ntpcal_split ds;
       int32_t      ts[3];
       uint32_t     uw, ud, sf32;

       /*
        * Split NTP time into days and seconds, shift days into CE
        * domain and process the parts.
        */
       ds = ntpcal_daysplit(ntp);

       /* split time part */
       ds.hi += priv_timesplit(ts, ds.lo);
       id->hour   = (uint8_t)ts[0];
       id->minute = (uint8_t)ts[1];
       id->second = (uint8_t)ts[2];

       /* split days into days and weeks, using floor division in unsigned */
       ds.hi += DAY_NTP_STARTS - 1; /* shift from NTP to RDN */
       sf32 = int32_sflag(ds.hi);
       ud   = (uint32_t)ds.hi;
       uw   = sf32 ^ ((sf32 ^ ud) / DAYSPERWEEK);
       ud  -= uw * DAYSPERWEEK;

       ds.hi = uint32_2cpl_to_int32(uw);
       ds.lo = ud;

       id->weekday = (uint8_t)ds.lo + 1;       /* weekday result    */

       /* get year and week in year */
       ds = isocal_split_eraweeks(ds.hi);      /* elapsed years&week*/
       id->year = (uint16_t)ds.hi + 1;         /* shift to current  */
       id->week = (uint8_t )ds.lo + 1;

       return (ds.hi >= 0 && ds.hi < 0x0000FFFF);
}

int
isocal_ntp_to_date(
       struct isodate *id,
       uint32_t        ntp,
       const time_t   *piv
       )
{
       vint64  ntp64;

       /*
        * Unfold ntp time around current time into NTP domain, then
        * convert the full time stamp.
        */
       ntp64 = ntpcal_ntp_to_ntp(ntp, piv);
       return isocal_ntp64_to_date(id, &ntp64);
}

/*
* Convert a ISO date spec into a second in the NTP time scale,
* properly truncated to 32 bit.
*/
vint64
isocal_date_to_ntp64(
       const struct isodate *id
       )
{
       int32_t weeks, days, secs;

       weeks = isocal_weeks_in_years((int32_t)id->year - 1)
             + (int32_t)id->week - 1;
       days = weeks * 7 + (int32_t)id->weekday;
       /* days is RDN of ISO date now */
       secs = ntpcal_etime_to_seconds(id->hour, id->minute, id->second);

       return ntpcal_dayjoin(days - DAY_NTP_STARTS, secs);
}

uint32_t
isocal_date_to_ntp(
       const struct isodate *id
       )
{
       /*
        * Get lower half of 64bit NTP timestamp from date/time.
        */
       return isocal_date_to_ntp64(id).d_s.lo;
}

/*
* ====================================================================
* 'basedate' support functions
* ====================================================================
*/

static int32_t s_baseday = NTP_TO_UNIX_DAYS;
static int32_t s_gpsweek = 0;

int32_t
basedate_eval_buildstamp(void)
{
       struct calendar jd;
       int32_t         ed;

       if (!ntpcal_get_build_date(&jd))
               return NTP_TO_UNIX_DAYS;

       /* The time zone of the build stamp is unspecified; we remove
        * one day to provide a certain slack. And in case somebody
        * fiddled with the system clock, we make sure we do not go
        * before the UNIX epoch (1970-01-01). It's probably not possible
        * to do this to the clock on most systems, but there are other
        * ways to tweak the build stamp.
        */
       jd.monthday -= 1;
       ed = ntpcal_date_to_rd(&jd) - DAY_NTP_STARTS;
       return (ed < NTP_TO_UNIX_DAYS) ? NTP_TO_UNIX_DAYS : ed;
}

int32_t
basedate_eval_string(
       const char * str
       )
{
       u_short y,m,d;
       u_long  ned;
       int     rc, nc;
       size_t  sl;

       sl = strlen(str);
       rc = sscanf(str, "%4hu-%2hu-%2hu%n", &y, &m, &d, &nc);
       if (rc == 3 && (size_t)nc == sl) {
               if (m >= 1 && m <= 12 && d >= 1 && d <= 31)
                       return ntpcal_edate_to_eradays(y-1, m-1, d)
                           - DAY_NTP_STARTS;
               goto buildstamp;
       }

       rc = sscanf(str, "%lu%n", &ned, &nc);
       if (rc == 1 && (size_t)nc == sl) {
               if (ned <= INT32_MAX)
                       return (int32_t)ned;
               goto buildstamp;
       }

 buildstamp:
       msyslog(LOG_WARNING,
               "basedate string \"%s\" invalid, build date substituted!",
               str);
       return basedate_eval_buildstamp();
}

uint32_t
basedate_get_day(void)
{
       return s_baseday;
}

int32_t
basedate_set_day(
       int32_t day
       )
{
       struct calendar jd;
       int32_t         retv;

       /* set NTP base date for NTP era unfolding */
       if (day < NTP_TO_UNIX_DAYS) {
               msyslog(LOG_WARNING,
                       "baseday_set_day: invalid day (%lu), UNIX epoch substituted",
                       (unsigned long)day);
               day = NTP_TO_UNIX_DAYS;
       }
       retv = s_baseday;
       s_baseday = day;
       ntpcal_rd_to_date(&jd, day + DAY_NTP_STARTS);
       msyslog(LOG_INFO, "basedate set to %04hu-%02hu-%02hu",
               jd.year, (u_short)jd.month, (u_short)jd.monthday);

       /* set GPS base week for GPS week unfolding */
       day = ntpcal_weekday_ge(day + DAY_NTP_STARTS, CAL_SUNDAY)
           - DAY_NTP_STARTS;
       if (day < NTP_TO_GPS_DAYS)
           day = NTP_TO_GPS_DAYS;
       s_gpsweek = (day - NTP_TO_GPS_DAYS) / DAYSPERWEEK;
       ntpcal_rd_to_date(&jd, day + DAY_NTP_STARTS);
       msyslog(LOG_INFO, "gps base set to %04hu-%02hu-%02hu (week %d)",
               jd.year, (u_short)jd.month, (u_short)jd.monthday, s_gpsweek);

       return retv;
}

time_t
basedate_get_eracenter(void)
{
       time_t retv;
       retv  = (time_t)(s_baseday - NTP_TO_UNIX_DAYS);
       retv *= SECSPERDAY;
       retv += (UINT32_C(1) << 31);
       return retv;
}

time_t
basedate_get_erabase(void)
{
       time_t retv;
       retv  = (time_t)(s_baseday - NTP_TO_UNIX_DAYS);
       retv *= SECSPERDAY;
       return retv;
}

uint32_t
basedate_get_gpsweek(void)
{
   return s_gpsweek;
}

uint32_t
basedate_expand_gpsweek(
   unsigned short weekno
   )
{
   /* We do a fast modulus expansion here. Since all quantities are
    * unsigned and we cannot go before the start of the GPS epoch
    * anyway, and since the truncated GPS week number is 10 bit, the
    * expansion becomes a simple sub/and/add sequence.
    */
   #if GPSWEEKS != 1024
   # error GPSWEEKS defined wrong -- should be 1024!
   #endif

   uint32_t diff;
   diff = ((uint32_t)weekno - s_gpsweek) & (GPSWEEKS - 1);
   return s_gpsweek + diff;
}

/*
* ====================================================================
* misc. helpers
* ====================================================================
*/

/* --------------------------------------------------------------------
* reconstruct the centrury from a truncated date and a day-of-week
*
* Given a date with truncated year (2-digit, 0..99) and a day-of-week
* from 1(Mon) to 7(Sun), recover the full year between 1900AD and 2300AD.
*/
int32_t
ntpcal_expand_century(
       uint32_t y,
       uint32_t m,
       uint32_t d,
       uint32_t wd)
{
       /* This algorithm is short but tricky... It's related to
        * Zeller's congruence, partially done backwards.
        *
        * A few facts to remember:
        *  1) The Gregorian calendar has a cycle of 400 years.
        *  2) The weekday of the 1st day of a century shifts by 5 days
        *     during a great cycle.
        *  3) For calendar math, a century starts with the 1st year,
        *     which is year 1, !not! zero.
        *
        * So we start with taking the weekday difference (mod 7)
        * between the truncated date (which is taken as an absolute
        * date in the 1st century in the proleptic calendar) and the
        * weekday given.
        *
        * When dividing this residual by 5, we obtain the number of
        * centuries to add to the base. But since the residual is (mod
        * 7), we have to make this an exact division by multiplication
        * with the modular inverse of 5 (mod 7), which is 3:
        *    3*5 === 1 (mod 7).
        *
        * If this yields a result of 4/5/6, the given date/day-of-week
        * combination is impossible, and we return zero as resulting
        * year to indicate failure.
        *
        * Then we remap the century to the range starting with year
        * 1900.
        */

       uint32_t c;

       /* check basic constraints */
       if ((y >= 100u) || (--m >= 12u) || (--d >= 31u))
               return 0;

       if ((m += 10u) >= 12u)          /* shift base to prev. March,1st */
               m -= 12u;
       else if (--y >= 100u)
               y += 100u;
       d += y + (y >> 2) + 2u;         /* year share */
       d += (m * 83u + 16u) >> 5;      /* month share */

       /* get (wd - d), shifted to positive value, and multiply with
        * 3(mod 7). (Exact division, see to comment)
        * Note: 1) d <= 184 at this point.
        *       2) 252 % 7 == 0, but 'wd' is off by one since we did
        *          '--d' above, so we add just 251 here!
        */
       c = u32mod7(3 * (251u + wd - d));
       if (c > 3u)
               return 0;

       if ((m > 9u) && (++y >= 100u)) {/* undo base shift */
               y -= 100u;
               c = (c + 1) & 3u;
       }
       y += (c * 100u);                /* combine into 1st cycle */
       y += (y < 300u) ? 2000 : 1600;  /* map to destination era */
       return (int)y;
}

char *
ntpcal_iso8601std(
       char *          buf,
       size_t          len,
       TcCivilDate *   cdp
       )
{
       if (!buf) {
               LIB_GETBUF(buf);
               len = LIB_BUFLENGTH;
       }
       if (len) {
               int slen = snprintf(buf, len, "%04u-%02u-%02uT%02u:%02u:%02u",
                              cdp->year, cdp->month, cdp->monthday,
                              cdp->hour, cdp->minute, cdp->second);
               if (slen < 0)
                       *buf = '\0';
       }
       return buf;
}

/* -*-EOF-*- */